A multilayer neural network has been trained by back propagation BP. This neural network adopts length, width, height and the number of parts in the mold as input and nine parameters (length, width, and height of up and down set-in, mold bases side thickness, bottom thickness of the core, and cavity plates) as output.